Founding Research Scientist - World Models & Self-Play

OMN4I; München; On-site; EUR 70,000 - 120,000; Full time

Job summary

OMN4I is building a physics-grounded world model as the shared brain layer for general-purpose robot fleets. You will own a large part of the research that decides whether this bet holds, shaping world-model architectures and self-play loops with real rollout data from our robot fleet (Unitree G1/B2W, ROSbot 3, Z1).

You will collaborate with academic partners and help define the research agenda as one of the first hires, reading robotics and physics simulation code and driving experiments.

Qualifications

• Strong research background in reinforcement learning, world models, or model-based control.
• Hands-on experience with self-play, model-based RL, or learned dynamics models.
• Comfortable owning an open-ended research problem in a founding-stage team.
• Ability to read and reason about robotics or physics simulation code.

Responsibilities

• Define and drive research on world-model architectures — dynamics models, physics priors, self-play curricula for robotic control.
• Design and run self-play training loops that scale and diagnose where they break.
• Own scaling experiments and ablations end-to-end, from hypothesis to written‑up result.
• Work directly with real rollout data from our robot fleet access to close the sim-to-real loop.
• Collaborate with our academic network (TUM, MIRMI) and publish where it makes sense.

Skills

RL research

World models

Model-based control

Self-play

Education

PhD or equivalent

Tools

SE(3) architectures

Job description

Mission

We're building a physics-grounded world model as the shared "brain layer" for general-purpose robot fleets — hardware-agnostic, cross-embodiment. The core bet: a learned world model that recalibrates itself from real rollouts plus physics priors, trained through self-play, beats approaches capped by human demonstration data. You'd own a large piece of the research that decides whether that bet holds.

What You'll Do

• Define and drive research on world-model architectures — dynamics models, physics priors, self-play curricula for robotic control.
• Design and run self-play training loops that scale, and diagnose where they break.
• Own scaling experiments and ablations end-to-end, from hypothesis to written‑up result.
• Work directly with real rollout data from our robot fleet access (Unitree G1/B2W, ROSbot 3, Z1) to close the sim-to-real loop.
• Collaborate with our academic network (TUM, MIRMI), publish where it makes sense, and help shape the research agenda as one of the first hires.

What We Look For

• Strong research background in reinforcement learning, world models, or model-based control — PhD or equivalent industry research experience.
• Hands‑on experience with self-play, model-based RL, or learned dynamics models (video‑generation and physics-informed learning backgrounds also welcome — we care more about depth than the exact subfield).
• Comfortable owning an open-ended research problem without a lot of hand-holding; this is a founding‑stage team, not an established lab.
• Can read and reason about robotics or physics simulation code, even if that's not your primary focus.

Nice to Have

• First-author publications on world models, model-based RL, self-play, or video/3D generative modeling.
• Experience with SE(3)-equivariant architectures or other structured/geometric priors.
• Track record of mentoring junior researchers or informally leading a small research effort.
• Prior sim-to-real experience on real robot hardware.